Fully Polarimetric Land Cover Classification Based on Markov Chains

Abstract

A novel \nland cover classification procedure is presented utilizing the information content of fully polarimetric SAR images. \nThe Cameron coherent target decomposition (CTD) is employed to \ncharacterize land cover pixel by pixel. Cameron’s CTD is employed since it \nprovides a complete set of elementary \nscattering mechanisms to describe the physical properties of the \nscatterer. The novelty of the proposed land classification approach lies on the \nfact that the features used for classification are not the types of the \nelementary scatterers themselves, but the \nway these types of scatterers alternate from pixel to pixel on the SAR image. Thus, transition \nmatrices that represent local Markov models are used as classification \nfeatures for land cover classification. The classification rule employs only \nthe most important transitions for decision making. The Frobenius inner product \nis employed as similarity measure. Ten different types of land cover are used \nfor testing the proposed method. In this aspect, the classification performance \nis significantly high.